search

sdss / lvmdrp

Local Volume Mapper (LVM) Data Reduction Pipeline
BSD 3-Clause "New" or "Revised" License
2 stars 0 forks source link

Tracing is very slow #92

Closed ndrory closed 1 week ago

ndrory commented 4 months ago

Investigate ways to make tracing faster. For example, fitting the profiles might be able to do with lower ftol and xtol values

ajmejia commented 2 months ago

As part of the improvements in the Gaussian fitting routines used in wavelength/LSF fitting (PR #129), we switched to method='dogbox' in scipy.optimize.least_squares (see commit 894a6dcf11aca5b7f363bd780aa5f742228ed407). This significantly speeds up tracing by a factor of >2: going from ~20-30min to ~10min per camera.

The consistency of the traces using one method or the other doesn't seem to be affected (TODO: verify this).

ndrory commented 1 week ago

as of sep 26, drp runtime is down to ~1200s at Utah. Extraction is ~10s and no longer dominant. The most computentially expensive step is now cosmic ray rejection.